Speak

Text-to-speech for AI coding assistants — hear responses spoken aloud using local TTS engines. Supports OpenCode, Claude Code, and Codex CLI. No cloud, no API keys, no subscriptions.

Features

• Multi-platform — works with OpenCode, Claude Code, and Codex CLI
• Two local TTS engines — Kokoro (82M params, 54 voices) and Supertonic 3 (99M params, 10 voices)
• 100% offline — no cloud APIs, no tokens, fully private
• On-demand — TTS is off by default, toggle on/off any time
• No background processes — models load per-request, zero idle RAM
• Shared config — settings sync across all platforms via ~/.config/opencode-speak/
• 54+ voices — multilingual: English, Japanese, Chinese, Hindi, French, Italian, Portuguese, Spanish, Korean

TTS Engine Comparison

Our pick: Kokoro for quality, Supertonic 3 for speed and language coverage. Both are CPU-only, zero-daemon, and load on-demand.

Install

Step 1: Install a TTS engine

You need at least one. Both are optional — the plugin auto-detects what's available.

# Kokoro (recommended) — 92MB model, 54 voices, CPU-optimized
uv tool install kokoro-cli
sudo apt install espeak-ng          # required on Linux
kokoro speak "hello" --play         # downloads model on first run

# Supertonic 3 — 400MB model, 10 voices, 31 languages
uv tool install speak-cli
speak "hello"                       # downloads model on first run
speak --stop                        # stop background daemon after first run

Step 2: Install for your platform

opencode plugin opencode-speak --global

npm install opencode-speak

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-speak"],
  "command": {
    "tts": {
      "template": "$ARGUMENTS",
      "description": "TTS: on | off | kokoro | speak | voice NAME | voices | test | help"
    }
  }
}

/plugin marketplace add ahmed0magdy/opencode-speak
/plugin install opencode-speak@opencode-speak

/opencode-speak:tts on       # start speaking responses
/opencode-speak:tts off      # stop, and cut off anything mid-sentence
/opencode-speak:tts toggle   # flip between on and off
/opencode-speak:tts voice af_bella

git clone https://github.com/ahmed0magdy/opencode-speak.git ~/.local/share/opencode-speak
mkdir -p ~/.codex/plugins
cp -r ~/.local/share/opencode-speak/codex ~/.codex/plugins/opencode-speak

tts-config.sh set enabled true

Step 3: Enable and test

# Enable TTS (shared across all platforms)
tts-config.sh set enabled true

# Test
tts-speak.sh --text "Hello, TTS is working!"

Usage

OpenCode (slash commands)

All platforms (config script)

# Enable/disable — `off` also stops speech already playing
tts-config.sh on
tts-config.sh off
tts-config.sh toggle
tts-config.sh stop        # stop current speech, stay enabled

# Switch engine
tts-config.sh set engine kokoro
tts-config.sh set engine speak

# Change voice
tts-config.sh set voice_kokoro af_bella
tts-config.sh set voice_speak emma

# Show status
tts-config.sh status

Settings live in ~/.config/opencode-speak/ (one file per key) and persist across sessions and restarts — turning TTS on once keeps it on until you turn it off. Set OPENCODE_SPEAK_CONFIG_DIR to point every component at a different directory, which is how the self-test runs without touching your real settings:

bash bin/tts-selftest.sh          # full run with audio, ~40s
bash bin/tts-selftest.sh --quiet  # config and plumbing checks only

Configuration (OpenCode only)

Pass options when using as an npm plugin:

{
  "plugin": [
    ["opencode-speak", {
      "defaultEngine": "kokoro",
      "defaultVoice": { "kokoro": "af_bella", "speak": "emma" },
      "maxChars": 3000,
      "autoStart": false
    }]
  ]
}

Voices

Kokoro — Female

af_heart af_bella af_nova af_sky af_jessica af_nicole af_aoede af_kore af_alloy af_river af_sarah

Kokoro — All 54 voices

English (American): af_*, am_* | English (British): bf_*, bm_* | Spanish: ef_*, em_* | French: ff_* | Hindi: hf_*, hm_* | Italian: if_*, im_* | Japanese: jf_*, jm_* | Portuguese: pf_*, pm_* | Chinese: zf_*, zm_*

Supertonic 3 — Female

sara emma lily maya nora

Supertonic 3 — Male

james daniel leo ryan noah

How It Works

Architecture

┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│   OpenCode   │    │  Claude Code │    │  Codex CLI   │
│  (TypeScript │    │  (Stop hook  │    │  (Stop hook  │
│   plugin)    │    │   + stdin)   │    │ + transcript)│
└──────┬───────┘    └──────┬───────┘    └──────┬───────┘
       │                   │                   │
       └───────────────────┴───────────────────┘
                                    │
                        ┌───────────▼───────────┐
                        │   bin/tts-speak.sh    │
                        │   (shared core)       │
                        └───────────┬───────────┘
                                    │
                        ┌───────────▼───────────┐
                        │ ~/.config/opencode-   │
                        │ speak/ (shared state) │
                        └───────────┬───────────┘
                                    │
                     ┌──────────────┼──────────────┐
                     ▼                             ▼
            ┌────────────────┐            ┌────────────────┐
            │  kokoro speak  │            │  speak --no-   │
            │  --service off │            │  daemon        │
            └────────────────┘            └────────────────┘

1. Each platform hooks into the "assistant finished responding" event
2. The hook extracts the last assistant message (method varies by platform)
3. Text is stripped of markdown formatting
4. Piped to the selected TTS engine CLI
5. Audio plays through system speakers

No background processes run while TTS is off. Models load on-demand and release memory after each synthesis.

Comparison with Other Tools

WSL2 Audio Setup

If you're on WSL2, make sure WSLg is enabled:

1. Edit C:\Users\<you>\.wslconfig:

   [wsl2]
   guiApplications=true
   

2. Restart WSL: wsl --shutdown from PowerShell
3. Verify: pactl info | grep "Server Name" should show PulseAudio (on PipeWire)

Uninstall

OpenCode

# Remove from opencode.jsonc: delete "opencode-speak" from "plugin" array and "tts" command
# Or remove local file:
rm ~/.config/opencode/plugins/opencode-speak.ts

Claude Code

rm -rf ~/.claude/plugins/opencode-speak

Codex CLI

rm -rf ~/.codex/plugins/opencode-speak

Shared (all platforms)

# Remove shared config
rm -rf ~/.config/opencode-speak

# Remove repo clone
rm -rf ~/.local/share/opencode-speak

# Remove TTS engines (optional)
uv tool uninstall kokoro-cli
uv tool uninstall speak-cli
rm -rf ~/.local/share/kokoro        # Kokoro models
rm -rf ~/.cache/supertonic3         # Supertonic 3 models

Dependencies

• jq — JSON parsing for bash hooks (Claude Code, Codex): sudo apt install jq
• kokoro-cli and/or speak-cli — TTS engines (see Install above)
• espeak-ng — required for Kokoro on Linux: sudo apt install espeak-ng

Contributing

Issues and pull requests welcome at github.com/ahmed0magdy/opencode-speak.

License

MIT